Google has released Magika, a machine-learning-based file identifier, to the public as part of its AI Cyber Defense Initiative. This initiative aims to provide improved automated tools for IT network defenders and other users.
Identifying the true content of a user-submitted file can be challenging. Simply relying on the file extension or using heuristics and human-crafted rules like libmagic to determine the file’s actual nature can be time-consuming and prone to errors, according to Google.
Magika, on the other hand, utilizes a trained model to quickly determine file types from file data. This approach is trusted by Google and is currently utilized in Gmail, Google Drive, Chrome’s Safe Browsing, and VirusTotal for accurate file identification and routing for further processing.
Google emphasizes the importance of using AI technology to enhance security measures, especially in the face of increasing automation of cyber intrusions and vulnerability research by malicious actors. Phil Venables, Google Cloud’s chief information security officer, and Royal Hansen, vice president of engineering for privacy, safety, and security, believe that Magika can empower network defenders to identify file content rapidly and at scale, thereby aiding in malware analysis and intrusion detection.
In the broader context of cybersecurity, AI models such as Magika can play a crucial role in inspecting files for suspicious content, identifying vulnerabilities in source code, and even generating patches to address bugs. Google’s engineers have also been exploring Gemini to enhance automated fuzzing of open source projects.
Google asserts that Magika outperforms its previous handcrafted rules system by being 50% more accurate in file type identification, taking milliseconds to classify a file type, and achieving at least 99% accuracy in tests. While not flawless, as it fails to classify file types about three percent of the time, Magika is licensed under Apache 2.0, with its code available for review.
Beyond Magika, Google plans to collaborate with 17 startups in the UK, US, and Europe through its AI Cyber Defense Initiative to train them in utilizing automated tools for enhanced security. Additionally, the company will expand its Cybersecurity Seminars Program with a \(15 million investment to support cybersecurity education in universities, and allocate \)2 million in grants for cyber-offense research and language model development at select academic institutions.
In conclusion, Google envisions AI as a key player in addressing cybersecurity challenges and paving the way for a safer and more trusted digital environment.
